<strong>Machine Learning (ML) technology for solar photovoltaic (PV) systems has emerged as a good option for increasing energy conversion efficiency under varying environmental conditions. This paper presents an adaptive Maximum Power Point Tracking (MPPT) approach using ML techniques to optimize real-time energy harvesting in PV systems. Traditional MPPT techniques such as Perturb and Observe (P&amp;O) and Incremental Conductance are less efficient under rapidly varying irradiance and temperature. But the proposed ML-based MPPT scheme learns dynamically from environment and system data and forecast and optimize the operating point with a better accuracy. Various supervised learning models are compared on simulated data to identify the most accurate model with respect to accuracy, convergence speed, and computational cost. Experimental validation by MATLAB/Simulink confirms the better performance of the adaptive ML-based MPPT approach compared to conventional approaches. This paper illustrates the potential of intelligent control in improving the robustness and efficiency of PV systems in real-world applications</strong>
The rapid growth of renewable energy sources, including wind power, has created a strong imperative for efficient and reliable infrastructure maintenance. This paper introduces a distributed system framework for UAV-based inspection of renewable energy facilities, with a focus on wind turbines. This framework combines the power of UAVs that have high resolution, thermal imaging, and visual sensors within a modular architecture for real-time collection, processing, and detection of defects. With the utilization of distributed computing and intelligent algorithms, this system reduces operational inefficiencies caused by the manual traditional method while enhancing the accuracy of inspections. An experimental study conducted at a wind power plant demonstrates the effectiveness of the system, achieving up to 94% defect detection accuracy while reducing inspection time per turbine to 1.5 hours. The proposed framework minimizes human intervention, enhances safety, and provides a cost-effective solution for monitoring large-scale renewable energy assets, contributing to the sustainability and reliability of modern energy infrastructure
Crowdsourcing has transformed conventional methods of problem-solving by gathering human intelligence on a significant scale, although the value of the method is often decreased, based on bias, inconsistency, and scalability. In this paper, we will present a conceptual and technical overview of how Artificial Intelligence (AI) and Machine Learning (ML) can help improve human-AI collaboration on crowdsourcing platforms and improve social computing. In AI&#39;s role as a complement to human response in crowdsourcing, by using unified ML algorithms to give out tasks and make decisions about quality, we expect AI will help produce more consistent, fluid, and scalable crowdsourced solutions. Some of the hybrid models of AI and human reasoning discussed in this paper include the following: AI that promotes optimal workflow, eliminates redundancy, and generally helps to improve the solidity of group decision-making. Ethical considerations discussed in this paper include removing bias, transparency, and building trust to name a few, to ensure AI usage in social computing systems will be more just and accountable<em>. </em>The results show that an adaptive AI infrastructure can significantly influence the efficiency and quality of the crowdsourced activities and leads to stronger and more sustainable social computing applications. The paper proposes an agenda for future research on human-AI collaboration, in the area of explanation and safety at the centred focus of AI-driven crowdsourcing contexts
Guaranteeing data privacy when working with machine learning is a difficult issue, especially in the federated learning or distributed learning settings. Differential privacy (DP) is often used as an approach to resolve this issue, which works by introducing noise into the training process. Excessive noise can, however, impair model performance. This paper investigates a different strategy by exploiting structured randomness in model parallelism and iteration subsampling to boost privacy without compromising accuracy and introduce a coherent framework that systematically combines model partitioning&mdash;where every client updates only part of model parameters&mdash;and balanced iteration subsampling&mdash;where points are involved in a constant number of rounds of training. Our analysis provides privacy amplification guarantees for both approaches, showing that these structured randomization methods lead to much better privacy than traditional Poisson subsampling or independent dropout techniques. This research also empirically verifies our solution on deep learning models, exhibiting better trade-offs between utility of the model and privacy protection. The proposed solution is able to diminish dependency on the high levels of noise by being a scalable and efficient privacy-preserving machine learning approach. The paper adds to the wider scope of secure AI through its unique contribution to the area of optimization in differential privacy, balancing the areas of privacy, efficiency of computation, and accuracy of the model
<strong>We introduce RaP-ProtoViT, an end-to-end dual-head transformer for 8-class GI endoscopy (Kvasir-v2). A margin head (ArcFace/AM-Softmax) enforces angular separation, while a prototype head aggregates top-k token&ndash;prototype similarities (with M trainable prototypes/class); a lightweight input-adaptive MLP fuses the heads. A leakage-aware pipeline (pHash dedup + GroupKFold) prevents near-duplicate bleed-over. Training uses AdamW(+SAM) with cosine warm-up, DropPath, label smoothing, SWA, and post-hoc temperature scaling; two-stage HPO (MOTPE+ASHA &rarr; qEHVI) under Latency@224 &le; 200 ms and memory constraints selects operating points. On Kvasir-v2 the model attains 99.1% accuracy, Macro-F1 = 0.991, Macro-AUPRC = 0.997, AUROC = 0.998, and ECE &asymp; 0.9%, with per-class F1 tightly clustered in 0.988&ndash;0.994 and fold stability (&plusmn;0.2 pp accuracy, &plusmn;0.002 Macro-F1). Ablations show margin-only/prototype-only variants reduce Macro-F1 to 0.967/0.975 and raise ECE to 2.8%/2.2%; removing adaptive fusion drops Macro-F1 to 0.984. The proposed HPO converges 2&ndash;3&times; faster and yields better final MF1/AUPRC/ECE than Bayesian TPE or Random+ASHA. The prototype head provides localized, intrinsically interpretable evidence, complementing the margin head&rsquo;s discrimination, within a single-model deployment footprint. By advancing robust, interpretable, and computationally efficient AI for gastric endoscopy, our approach can improve early detection of gastrointestinal disease and enable reliable clinical deployment across diverse healthcare settings.</strong>
Predicting currency exchange rates, especially volatile pairs like GBP/USD, is challenging because prices depend on many interacting economic, political, and market factors. Traditional forecasting approaches often struggle with the nonlinear, non-stationary behavior of financial time series. The paper proposes a Multi-Layer Perceptron (MLP) whose hyperparameters are optimized with Particle Swarm Optimization (PSO). PSO automatically searches the hyperparameter space, replacing slow manual tuning and finding better network configurations. Experiments show the PSO-optimized MLP reduces RMSE by 45.33\% relative to a manually tuned baseline, indicating markedly improved predictive accuracy under market volatility. The study demonstrates that swarm-intelligence optimization is an effective, repeatable way to build stronger neural forecasts for foreign exchange. By improving forecasting reliability, the work supports SDG 8 and SDG 9 through smarter, AI-driven financial decision support. Swarm intelligence proves practical for robust forex forecasting. Such models can assist traders, firms, and regulators in risk management and efficient currency operations.
IoT security needs to be strengthened as smart homes are more connected and prone to complex cyberattacks. One innovative approach to identify and lower negative behavior in IoT networks is prestige-based connection monitoring. However, by having restricted scalability, poor trust assessment, and vulnerability to node impersonation or penetration, existing methods might threaten the integrity of the network. This work proposes a Blockchain-based Reputation System (BC-RS) based dynamic and transparent node trust assessment to tackle these limits in order to avoid hostile node infiltration in smart home networks. By use of distributed ledgers, the BC-RS system assures tamper-proof integrity of reputation data, computes trust ratings, and records node interactions. The proposed method monitors node behavior, finds real-time harmful activity, and prevents untrusted nodes from network connection. Experimental analysis shows that the framework enhances network security by increasing trust score accuracy by 98.2%, reducing false positives in threat detection by 39%, and improving system resilience by 97.6%. These findings confirm the effectiveness of prestige-based tracking combined with blockchain in strengthening IoT security and ensuring safer smart home environments.
The collection of data significantly influences generalizability, reliability, and model accuracy in key aspects of machine learning (ML). This work presents a thorough review of many data collection methods along with their challenges in effective use. Current methods compromise the quality of ML outputs by means of bias, inconsistency, scalability limits, and lack of standardization. Data gathering strategies are selected depending on Taxonomy Analysis with ML (TA-ML), thereby addressing these issues. Based on intended use, data type, source dependability, and collection size, the framework arranges methods. This rigorous approach helps practitioners to choose appropriate strategies suited for the conditions of their assignment. By means of the recommended strategy, users will be able to reduce noise, enhance data relevance, and reduce bias, thus increasing model performance. Moreover highlighted in the study is how numerous ML disciplines&#39; structure helps sensible decision-making. Results reveal that the proposed taxonomy-based strategy properly addresses normal data collection issues and helps more accurate and efficient ML development. Reaching the decision-making mark with 98.32% accuracy by 97.6%, efficiency by 96.3%, the recommended strategy is evident.
In the data-driven economy of today, artificial intelligence (AI) enables speedy analysis of large datasets, therefore revealing valuable insights that direct strategic choices. In many sectors, this approach accelerates the recognition of trends, deviations, and patterns. However, traditional data analysis methods are often slow, entail tremendous human work, and struggle with scalability when faced with high-volume, high-velocity data. This study offers a fresh perspective to overcome these limitations: Fast Trend Discovery and Insight Extraction from Business or Social Data Applied using AutoML Tools (AMLT). Modern AutoML technologies used here automates the end-to- end data analysis pipeline&mdash;data preparation, model selection, and insight generation. The proposed method finds user behavior patterns, market trends, and attitude changes by use of real-world data from business intelligence systems and social media analytics. Results reveal that AMLT accelerates decision-making compared to manual analysis techniques, reduces analysis time by more than 60%, and increases model consistency. The framework looks to be fairly useful for non-technical users especially as it offers scalability insight generating. The proposed method achieves the data analysis time by 35%, model accuracy by 97.4%, decision making by 98.3%.
Machine learning has emerged as a transformative tool in the medical field, particularly in enhancing diagnostic accuracy and therapeutic decision-making. In the context of gastrointestinal (GI) diseases, its application is reshaping early detection and treatment strategies. Traditional GI diagnostic methods often rely heavily on manual interpretation of endoscopic images, which can be time-consuming and subject to inter-observer variability. This can lead to delays in diagnosis and inconsistent therapeutic outcomes. To address these limitations, we propose a Convolutional Neural Network-based system for Analyzing Endoscopic Images (CNN-AEI), aimed at improving the early detection of gastrointestinal abnormalities. This system automates image analysis using deep learning, enabling real-time assessment with higher precision. The proposed method is implemented to support clinicians by providing accurate, consistent, and rapid diagnostic feedback from endoscopic imagery. Experimental results demonstrate that the CNN-AEI framework significantly improves diagnostic accuracy, sensitivity, and specificity compared to conventional assessment methods. This advancement has the potential to reduce diagnostic errors and support timely therapeutic interventions, ultimately enhancing patient outcomes in GI care.
From the gaming to leisure industries, augmented reality (AR) is providing more dynamic and interactive experiences. The integration of 5G technology amplifies AR&rsquo;s potential due to ultra-reliable, extremely low-latency, high-speed connections in real-time applications. Current AR gaming systems, particularly in multiplayer settings, face challenges such as lag, limited bandwidth, and heavy traffic on networks. These issues interfere with real-time communication and interaction resulting in a negative experience. To attempt to resolve these issues, Real-Time Multiplayer AR Gaming Experience exploiting Edge Computing Integration with 5G Networks (EC-I-5GN) has been proposed as a solution placing Edge Computing Integration with 5G Networks second. The integration of edge computing with 5G networks facilitates the processing of data closer to the point of use, lowering latency and improving bandwidth utilization. The architecture improves multiplayer AR games by providing better gameplay and reducing disruptions making these games truly interactive. The results suggest that the use of the EC-I-5GN framework improves the real-time responsiveness, interaction with the game, and overall satisfaction while addressing the key limitations of current AR gaming technologies.
Medical image analysis is an important task in diagnosing various diseases, among which the study of megaloblastic anemia stands out resulting in improving the publish health. The peculiarity of processing the corresponding images is the selection of the edge for all objects of interest, including the details of the structure of megaloblastic anemia cells. It is shown that for these purposes it is advisable to use the ideology of wavelets. Based on this, the paper considers the issues of the influence of the wavelet type on the selection of the edge in images with megaloblastic anemia cells with changed contrast in the RGB and HSV color spaces. The wavelets considered in this paper include gaus1, haar, db2, and bior1.1. For the purpose of comparing the results, such quality assessments as niqe, brisque, and entropy are used, as well as a visual comparison of the obtained results. It is shown that for selecting the edge and potential areas of interest in images with megaloblastic anemia cells, it is advisable to use the RGB space. It is also noted that the gaus1 wavelet allows for efficient allocation of potential areas of interest, while the haar and bior1.1 wavelets allow for allocation of the edges of objects of interest. The results are presented in the form of various figures and tables.
